In this study, we propose Mixed and Masked Image Modeling (MixMIM), a simple but efficient MIM method that is applicable to various hierarchical Vision Transformers. Existing MIM methods replace a random subset of input tokens with a special MASK symbol and aim at reconstructing original image tokens from the corrupted image. However, we find that using the MASK symbol greatly slows down the training and causes training-finetuning inconsistency, due to the large masking ratio (e.g., 40% in BEiT). In contrast, we replace the masked tokens of one image with visible tokens of another image, i.e., creating a mixed image. We then conduct dual reconstruction to reconstruct the original two images from the mixed input, which significantly improves efficiency. While MixMIM can be applied to various architectures, this paper explores a simpler but stronger hierarchical Transformer, and scales with MixMIM-B, -L, and -H. Empirical results demonstrate that MixMIM can learn high-quality visual representations efficiently. Notably, MixMIM-B with 88M parameters achieves 85.1% top-1 accuracy on ImageNet-1K by pretraining for 600 epochs, setting a new record for neural networks with comparable model sizes (e.g., ViT-B) among MIM methods. Besides, its transferring performances on the other 6 datasets show MixMIM has better FLOPs / performance tradeoff than previous MIM methods. Code is available at https://github.com/Sense-X/MixMIM.
Pedestrian dead reckoning is a challenging task due to the low-cost inertial sensor error accumulation. Recent research has shown that deep learning methods can achieve impressive performance in handling this issue. In this letter, we propose inertial odometry using a deep learning-based velocity estimation method. The deep neural network based on Res2Net modules and two convolutional block attention modules is leveraged to restore the potential connection between the horizontal velocity vector and raw inertial data from a smartphone. Our network is trained using only fifty percent of the public inertial odometry dataset (RoNIN) data. Then, it is validated on the RoNIN testing dataset and another public inertial odometry dataset (OXIOD). Compared with the traditional step-length and heading system-based algorithm, our approach decreases the absolute translation error (ATE) by 76%-86%. In addition, compared with the state-of-the-art deep learning method (RoNIN), our method improves its ATE by 6%-31.4%.
Ultrasonography is an important routine examination for breast cancer diagnosis, due to its non-invasive, radiation-free and low-cost properties. However, it is still not the first-line screening test for breast cancer due to its inherent limitations. It would be a tremendous success if we can precisely diagnose breast cancer by breast ultrasound images (BUS). Many learning-based computer-aided diagnostic methods have been proposed to achieve breast cancer diagnosis/lesion classification. However, most of them require a pre-define ROI and then classify the lesion inside the ROI. Conventional classification backbones, such as VGG16 and ResNet50, can achieve promising classification results with no ROI requirement. But these models lack interpretability, thus restricting their use in clinical practice. In this study, we propose a novel ROI-free model for breast cancer diagnosis in ultrasound images with interpretable feature representations. We leverage the anatomical prior knowledge that malignant and benign tumors have different spatial relationships between different tissue layers, and propose a HoVer-Transformer to formulate this prior knowledge. The proposed HoVer-Trans block extracts the inter- and intra-layer spatial information horizontally and vertically. We conduct and release an open dataset GDPH&GYFYY for breast cancer diagnosis in BUS. The proposed model is evaluated in three datasets by comparing with four CNN-based models and two vision transformer models via a five-fold cross validation. It achieves state-of-the-art classification performance with the best model interpretability.
The vision-based grasp detection method is an important research direction in the field of robotics. However, due to the rectangle metric of the grasp detection rectangle's limitation, a false-positive grasp occurs, resulting in the failure of the real-world robot grasp task. In this paper, we propose a novel generative convolutional neural network model to improve the accuracy and robustness of robot grasp detection in real-world scenes. First, a Gaussian-based guided training method is used to encode the quality of the grasp point and grasp angle in the grasp pose, highlighting the highest-quality grasp point position and grasp angle and reducing the generation of false-positive grasps. Simultaneously, deformable convolution is used to obtain the shape features of the object in order to guide the subsequent network to the position. Furthermore, a global-local feature fusion method is introduced in order to efficiently obtain finer features during the feature reconstruction stage, allowing the network to focus on the features of the grasped objects. On the Cornell Grasping Datasets and Jacquard Datasets, our method achieves excellent performance of 99.0$\%$ and 95.9$\%$, respectively. Finally, the proposed method is put to the test in a real-world robot grasping scenario.
Electro-optic (EO) modulation is a well-known and essential topic in the field of communications and sensing. Its ultrahigh efficiency is unprecedentedly desired in the current green and data era. However, dramatically increasing the modulation efficiency is difficult due to the monotonic mapping relationship between the electrical signal and modulated optical signal. Here, a new mechanism termed phase-transition EO modulation is revealed from the reciprocal transition between two distinct phase planes arising from the bifurcation. Remarkably, a monolithically integrated mode-locked laser (MLL) is implemented as a prototype. A 24.8-GHz radio-frequency signal is generated and modulated, achieving a modulation energy efficiency of 3.06 fJ/bit improved by about four orders of magnitude and a contrast ratio exceeding 50 dB. Thus, MLL-based phase-transition EO modulation is characterised by ultrahigh modulation efficiency and ultrahigh contrast ratio, as experimentally proved in radio-over-fibre and underwater acoustic-sensing systems. This phase-transition EO modulation opens a new avenue for green communication and ubiquitous connections.
Recommender systems have been successfully used in many domains with the help of machine learning algorithms. However, such applications tend to use multi-dimensional user data, which has raised widespread concerns about the breach of users privacy. Meanwhile, wearable technologies have enabled users to collect fitness-related data through embedded sensors to monitor their conditions or achieve personalized fitness goals. In this paper, we propose a novel privacy-aware personalized fitness recommender system. We introduce a multi-level deep learning framework that learns important features from a large-scale real fitness dataset that is collected from wearable IoT devices to derive intelligent fitness recommendations. Unlike most existing approaches, our approach achieves personalization by inferring the fitness characteristics of users from sensory data and thus minimizing the need for explicitly collecting user identity or biometric information, such as name, age, height, weight. In particular, our proposed models and algorithms predict (a) personalized exercise distance recommendations to help users to achieve target calories, (b) personalized speed sequence recommendations to adjust exercise speed given the nature of the exercise and the chosen route, and (c) personalized heart rate sequence to guide the user of the potential health status for future exercises. Our experimental evaluation on a real-world Fitbit dataset demonstrated high accuracy in predicting exercise distance, speed sequence, and heart rate sequence compared to similar studies. Furthermore, our approach is novel compared to existing studies as it does not require collecting and using users sensitive information, and thus it preserves the users privacy.
Purpose: To introduce a dual-domain reconstruction network with V-Net and K-Net for accurate MR image reconstruction from undersampled k-space data. Methods: Most state-of-the-art reconstruction methods apply U-Net or cascaded U-Nets in image domain and/or k-space domain. Nevertheless, these methods have following problems: (1) Directly applying U-Net in k-space domain is not optimal for extracting features in k-space domain; (2) Classical image-domain oriented U-Net is heavy-weight and hence is inefficient to be cascaded many times for yielding good reconstruction accuracy; (3) Classical image-domain oriented U-Net does not fully make use information of encoder network for extracting features in decoder network; and (4) Existing methods are ineffective in simultaneously extracting and fusing features in image domain and its dual k-space domain. To tackle these problems, we propose in this paper (1) an image-domain encoder-decoder sub-network called V-Net which is more light-weight for cascading and effective in fully utilizing features in the encoder for decoding, (2) a k-space domain sub-network called K-Net which is more suitable for extracting hierarchical features in k-space domain, and (3) a dual-domain reconstruction network where V-Nets and K-Nets are parallelly and effectively combined and cascaded. Results: Extensive experimental results on the challenging fastMRI dataset demonstrate that the proposed KV-Net can reconstruct high-quality images and outperform current state-of-the-art approaches with fewer parameters. Conclusions: To reconstruct images effectively and efficiently from incomplete k-space data, we have presented a parallel dual-domain KV-Net to combine K-Nets and V-Nets. The KV-Net is more lightweight than state-of-the-art methods but achieves better reconstruction performance.
Recent studies pointed out that knowledge distillation (KD) suffers from two degradation problems, the teacher-student gap and the incompatibility with strong data augmentations, making it not applicable to training state-of-the-art models, which are trained with advanced augmentations. However, we observe that a key factor, i.e., the temperatures in the softmax functions for generating probabilities of both the teacher and student models, was mostly overlooked in previous methods. With properly tuned temperatures, such degradation problems of KD can be much mitigated. However, instead of relying on a naive grid search, which shows poor transferability, we propose Meta Knowledge Distillation (MKD) to meta-learn the distillation with learnable meta temperature parameters. The meta parameters are adaptively adjusted during training according to the gradients of the learning objective. We validate that MKD is robust to different dataset scales, different teacher/student architectures, and different types of data augmentation. With MKD, we achieve the best performance with popular ViT architectures among compared methods that use only ImageNet-1K as training data, ranging from tiny to large models. With ViT-L, we achieve 86.5% with 600 epochs of training, 0.6% better than MAE that trains for 1,650 epochs.
It is a challenging task to learn rich and multi-scale spatiotemporal semantics from high-dimensional videos, due to large local redundancy and complex global dependency between video frames. The recent advances in this research have been mainly driven by 3D convolutional neural networks and vision transformers. Although 3D convolution can efficiently aggregate local context to suppress local redundancy from a small 3D neighborhood, it lacks the capability to capture global dependency because of the limited receptive field. Alternatively, vision transformers can effectively capture long-range dependency by self-attention mechanism, while having the limitation on reducing local redundancy with blind similarity comparison among all the tokens in each layer. Based on these observations, we propose a novel Unified transFormer (UniFormer) which seamlessly integrates merits of 3D convolution and spatiotemporal self-attention in a concise transformer format, and achieves a preferable balance between computation and accuracy. Different from traditional transformers, our relation aggregator can tackle both spatiotemporal redundancy and dependency, by learning local and global token affinity respectively in shallow and deep layers. We conduct extensive experiments on the popular video benchmarks, e.g., Kinetics-400, Kinetics-600, and Something-Something V1&V2. With only ImageNet-1K pretraining, our UniFormer achieves 82.9%/84.8% top-1 accuracy on Kinetics-400/Kinetics-600, while requiring 10x fewer GFLOPs than other state-of-the-art methods. For Something-Something V1 and V2, our UniFormer achieves new state-of-the-art performances of 60.9% and 71.2% top-1 accuracy respectively. Code is available at https://github.com/Sense-X/UniFormer.
Kalman filter-based Inertial Navigation System (INS) is a reliable and efficient method to estimate the position of a pedestrian indoors. Classical INS-based methodology which is called IEZ (INS-EKF-ZUPT) makes use of an Extended Kalman Filter (EKF), a Zero velocity UPdaTing (ZUPT) to calculate the position and attitude of a person. However, heading error which is a key factor of the whole Pedestrian Dead Reckoning (PDR) system is unobservable for IEZ-based PDR system. To minimize the error, Electronic Com-pass (EC) algorithm becomes a valid method. But magnetic disturbance may have a big negative effect on it. In this paper, the Quasi-static Magnetic field Detection (QMD) method is proposed to detect the pure magnetic field and then selects EC algorithm or Heuristic heading Drift Reduction algorithm (HDR) according to the detection result, which implements the complementation of the two methods. Meanwhile, the QMD, EC, and HDR algorithms are integrated into the IEZ framework to form a new PDR solution which is named Advanced IEZ (AIEZ).